Andrej Karpathy says only 5,000 people have seen what AI can really do
Short summary
Andrej Karpathy argues that only a tiny fraction of people—roughly 5,000—have seen what frontier AI systems can really accomplish, while most judge AI by casual chatbot interactions. The gap between capability and deployment is widening, and companies like Box are positioning themselves to bridge it by building enterprise applications on top of these models.
- Karpathy's breakdown: 6 billion people barely use LLMs, 1-2 billion use chatbots casually, 20 million use frontier models professionally, and only 5,000 have access to cutting-edge systems inside labs.
- AI capability peaks in domains that are digital, verifiable, and economically valuable—coding and math see the fastest improvements because results can be tested and measured, while writing and advice remain harder to evaluate.
- The real gains appear when agents work for hours on complex problems (like Karpathy's AutoResearch running 50 experiments overnight), not when answering single questions—most people don't stress these systems enough to see their actual potential.
- Box and other enterprise-focused companies see opportunity in helping organizations deploy AI agents at scale, though obstacles remain: unequal model capabilities, high costs, immature tooling, and the need for significant change management.
Full summary
Andrej Karpathy, now at Anthropic, argues there's a massive gap between what AI can actually do and what most people think it can do. By his estimate, roughly 6 billion people barely use language models at all, 1-2 billion use chatbots casually, about 20 million use frontier models professionally for coding and math, and only around 5,000 people inside AI labs have seen what these systems can really accomplish. Box CEO Aaron Levie agrees we're still in the early stages of AI adoption and points out that most people still view AI as just a chatbot or search engine replacement, missing the bigger picture entirely.
The real opportunity, according to both Karpathy and Levie, lies in how AI excels at work that's digital, verifiable, and economically valuable—like coding and mathematical problems where results can be tested and measured. Karpathy's own experience shows the shift: less than a year ago he was writing code by hand, but by April he was using agents that could tackle programming problems taking days or weeks in a single session. His AutoResearch project even ran 50 experiments overnight on one GPU, found a better learning rate, and committed it to Git. Writing and advice are harder for AI to improve at because there's no clear way to measure success, so the money and focus have concentrated on technical domains instead.
Box is betting on "agent swarms"—multiple AI agents working in parallel on complex tasks—as the next form factor for enterprise work. The obstacles are real though: models aren't equally capable across all domains, running them at scale is expensive, and companies still need to figure out data connectivity, security, compliance, and change management. Levie expects huge friction before industries outside software and AI research adopt this stuff. That's where Box sees its opening—building the applied layer on top of frontier models, handling the industry-specific infrastructure and workflows rather than competing with model providers directly.